Quality of Service Profiles in Web Service Discovery

Standardization of the description and delivery of XML-based web services has opened up a market, larger than any before, in ‘commercial off-the-shelf’ (COTS) software components. As a result, standardization efforts are being made towards the assembly of systems from web services where the coordination is defined by workflow languages. With several potential implementations for many of the tasks within such a system, the need has never been greater to aid, through automation, the discovery process. Furthermore when many functional equivalents exist, it becomes of great importance to discriminate between these on the basis of cost and performance. Cardoso and Sheth proposed a useful set of Quality of Service measurements [1] and a framework to apply such considerations within web service discovery [2] [3]. Unfortunately the ‘fitness’ metric defined in [2] contains mathematical flaws, which have been propagated to [3] and other publications. In particular: • Metrics are defined, and claimed normalized, but these can take negative values (and even undefined values due to division by zero); • Combinations of these are defined using the geometric mean (even though this doesn’t fit their informal claims) and so combined metrics can also be undefined; • Underand over-performance are not distinguished and are equally penalized. In proposing corrections for these problems [4] we have found alternative solutions to the latter issue that accommodate different design strategies. We present here the resulting system as an advance in the technique.